Scribd
Upload
3 views

Computer Vision

Computer vision aims to extract valuable insights from visual data, enabling applications such as facial recognition, autonomous vehicle navigation, and medical image analysis. Key challenges include converting physical signals into useful information and accurately interpreting 3D environments from 2D images. CCD technology plays a crucial role in image capture and processing, while various image formats and descriptors enhance the analysis of texture, color, and depth in visual data.

Uploaded by

naveen.a
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
3 views

Computer Vision

Computer vision aims to extract valuable insights from visual data, enabling applications such as facial recognition, autonomous vehicle navigation, and medical image analysis. Key challenges include converting physical signals into useful information and accurately interpreting 3D environments from 2D images. CCD technology plays a crucial role in image capture and processing, while various image formats and descriptors enhance the analysis of texture, color, and depth in visual data.

Uploaded by

naveen.a
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 3

Computer Vision

Aim of computer vision is to generate useful information and insights from scenery, sequences and
the objects, received from video cameras. To analyze patterns.

 Automatic face recognition and interpretation of expressions.


 visual guidance of autonomous vehicles
 automated medical image analysis, interpretation, and diagnosis
 robotic manufacturing: manipulation, grading, and assembly of parts
 OCR: recognition of printed or handwritten characters and words
 agricultural robots: visual grading and harvesting of produce
 smart offices: tracking of persons and objects; understanding gestures
 biometric-based visual identification of persons
 visually endowed robotic helpers
 security monitoring and alerting; detection of anomaly
 intelligent interpretive prostheses for the blind
 tracking of moving objects; collision avoidance; stereoscopic depth
 object-based (model-based) compression of video streams
 general scene understanding
Challenges:
Signal to symbol convertor (We are presented with physical signals in external environment
which is presented on sensory surface like retina, video camera, microphone. These signals must be
converted in order to get useful information and have better understanding on the environment at
play by the machine. The physical signals offer very less information about the environment as a
whole.

An image is a 2D optical projection but the world we wish to analyze or make sense is a 3D, so we
have to do inverse optics by converting this 3D projection to a 2D projection to get properties from
the image. But a 2D- 3D projection is mathematically impossible.
It is also inverse graphics. Graphics begins with 3D world description. Facial recognition presents huge
challenges in Computer vision. Humans can process and identify faces easily, machines often are
handicapped by textured backgrounds, colors, lightings
There needs to have a mathematical model which can
 Perform the figure ground segmentation for objects and background.
 Infer the 3D arrangements of the objects.
 Infer surface properties from 2D image statistics.
 Infer volumetric properties from 2D image properties.
 And all these computing has to be done in real time.

CCD (Charge Coupled Cameras)


CCD converts light into electrical charge and transfers that charge through a controlled sequence,
converts into digital signals and process it into a final image. CCD has array of sensors which has light
sensitive cells called pixels. Pixels capture the photons and convert them into electric charge
The process flows as follows
Photon Absorption: When light hits the surface of the CCD sensor, it energizes electrons within the
semiconductor material of each pixel, thereby breaking the electrons free from the atoms.

Charge Accumulation: The freed electrons collect in a potential bucket inside the pixel. The amount of
charge in each pixel corresponds to the intensity of light that fell on it. Bright areas of the scene
generate more electrons, and dark areas generate fewer electrons.

Shift Registers: The CCD sensor is designed with a series of shift registers, which are like conveyor
belts for electric charge. These registers move the accumulated charge from pixel to pixel in a
controlled sequence.

Analog-to-Digital Conversion: There, the analog charge signal is converted into a digital signal, which
can be stored, processed, and displayed by electronic devices like computers or screens.

Signal Processing: Once the digital signal is obtained, it can undergo various forms of processing, such
as noise reduction, color interpolation (for color imaging), and other adjustments to enhance the final
image quality.

Readout and Reset: After the charge from all the pixels has been read out, the CCD sensor needs to be
reset. This involves clearing the accumulated charge from each pixel, preparing the sensor for the next
exposure.

Formats

Jpeg – Ideal for variable compression of continuous images, DCT compression from 100:1 to 10:1

Mpeg- stream oriented used mainly for videos individual frames are jpeg compressed

Gif- ideal for sparsed binarized images, ideal for low bandwidth browsers, provides high compression

Tiff- non-compressive, Tagged image file formats, 24bit color

Bmp- non-compressive bit mapped format, individual pixel values can be easily extracted.

Texture color stereo and motion descriptors:

Common structure flow is ill posed insoluble problems of inference and raw data and convert into well
posed problems in which we can compute object properties.

Texture information:

Helps in object and scene identification. Also helps to identify the surface shape and in image
segmentation. Image segmentation converts images into collections of regions that are labelled. It can
help in easier identification and processing of the images. Texture can be defined by the existing
correlation across the images

Color information:

Helps in object and scene identification just like texture information but has difficulty in indetifying
wavelength. For eg when a light source is used to illuminate the object human eye can still process the
color of the object but computer vision has issue in processing the natural color when illuminated with
light. Retinex algorithm helps to solve this color constancy
Stereo information:

Information regarding the depth can be obtained by using two or more cameras. By increasing the
distance between eyes/cameras we can increase the depth of field of vision

You might also like